Skip to content

GitOps Principles

Overview

GitOps is a way of running systems where Git is the single source of truth for what should be deployed. Rather than pushing changes into a cluster, you describe the desired state in Git, and an agent makes reality match it.

The core principles

  • Declarative: the whole system is described as declarative files (what you want), not scripts (how to do it).
  • Versioned in Git: those files live in Git, so the desired state is versioned, reviewable, and auditable. Git history is a complete record of every change.
  • Pulled, not pushed: an agent inside the cluster pulls the desired state from Git, rather than an outside system pushing into the cluster.
  • Continuously reconciled: the agent constantly compares the cluster to Git and corrects any difference, automatically.

Why pull instead of push

The pull model is a deliberate choice with real benefits:

  • Security: no external system holds credentials to change the cluster. The agent only needs to read Git.
  • Continuous: a push happens once, at deploy time. A pulling agent runs always, so it also heals drift, not just deploys.

Why teams adopt it

Because every change goes through Git, GitOps gives you review, history, and easy rollback for free: reverting a deployment is reverting a commit. Combined with continuous reconciliation, the cluster becomes predictable and self-correcting, which is what makes large systems manageable.

Next